A splitting theorem for the Seiberg-Witten invariant of a homology \(S^1\times S^3\) (Q1637315)

Let \(X\) be a smooth oriented spin 4-manifold with the rational homology of \(S^1\times S^3\). \(X\) has two topological invariants \(\lambda_{\mathrm{SW}}(X)\) and \(h(X)\). \(\lambda_{\mathrm{SW}}\) is the Seiberg-Witten invariant with correction ([\textit{T. Mrowka} et al., J. Differ. Geom. 88, No. 2, 333--377 (2011; Zbl 1238.57028)], which guarantees invariance of \(\lambda_{\mathrm{SW}}\) to the choice of metric) with extension in this paper. \(h(X)\) is the invariant found in [\textit{K. A. Frøyshov}, Duke Math. J. 155, No. 3, 519--576 (2010; Zbl 1237.57033)]. Originally \(h(X)\) is defined by using a spin structure \(\mathfrak{s}\) and denoted by \(h(X;\mathfrak{s})\). As to the relation between these two invariants, the following theorem is proved in this paper, as the main theorem. \textbf{Theorem A}. Assume \(X\) is homology oriented by \(\alpha\in H^1(X;\mathbb{Z})\) and the Poincaré dual of \(\alpha\) is realized by a rational homology sphere \(Y\subset X\). Let \(\mathfrak{s}\) be the induced spin structure on \(Y\), and let \(W\) be the spin cobordism from \(Y\) to itself obtained by cutting \(X\) open along \(Y\). Then \[ \lambda_{\mathrm{SW}}(X)+h(X)=-\operatorname{Lef}(W_\ast:HM^{\mathrm{red}}(Y,\mathfrak{s})\to HM^{\mathrm{red}}(Y,\mathfrak{s})). \tag{1} \] This theorem provides an alternative proof of \textbf{Theorem B}. Under the same assumptions and notations, \(X\) admits no Riemannian metric of positive scalar curvature unless \(\lambda_{\mathrm{SW}}(X)+h(Y;\mathfrak{s})=0\). The authors remark that this theorem has already been proved in [\textit{J. Lin}, ``The Seiberg-Witten equations on end-periodic manifolds and an obstruction to positive scalar curvature metrics'', Preprint, \url{arXiv:1603.03698}], by different methods. The proof of Theorem B from Theorem A is given in \S4. Another application of Theorem A is to the study of homology (resp. \(\mathbb{Z}/2\) homology) cobordism. In dimension three, the homology cobordism, and the \(\mathbb{Z}/2\) homology cobordism groups are denoted by \(\Theta^3_\mathbb{Z}\) and \(\Theta^3_{\mathbb{Z}/2}\), respectively. The Rochlin invariant provides a surjective homomorphism \(\rho:\Theta^3_\mathbb{Z}\to \mathbb{Z}/2\). It has been conjectured that \(\rho(Y)=1\) implies that \(Y\) has infinite order in \(\Theta^3_\mathbb{Z}\). As an application of Theorem A, the following partial answer of this conjecture is given. \textbf{Theorem C}. Any \(h\)-positive or \(h\)-negative integral homology sphere \(Y\) with \(\rho(Y)=1\) has infinite order in \(\Theta^3_\mathbb{Z}\). Here, a \(\mathbb{Z}/2\) homology sphere is said to \(h\)-positive (resp. \(h\)-negative), if the reduced monopole homology \(\mathrm{M}^{\mathrm{red}}(Y,\mathfrak{s})\) corresponding to the unique spin structure \(\mathfrak{s}\) on \(Y\) is supported in degree \(\ge 2-2h(Y,\mathfrak{s})\) (resp. \(\le -2h(Y,\mathfrak{s})-1)\) (cf. Definition 5.3). As for the group \(\Theta^3_{\mathbb{Z}/2}\), denoting by \(\Theta^3_L\) the subgroup of \(\Theta^3_{\mathbb{Z}/2}\) generated by all the \(L\)-spaces over \(\mathbb{Q}\) that are \(\mathbb{Z}/2\) homology spheres, the following theorem is proved. \textbf{Theorem D}. Any \(h\)-positive or \(h\)-negative \(\mathbb{Z}/2\) homology sphere \(Y\) with \(\rho(Y)\not=h(Y)\) (mod 2) has infinite order in the group \(\Theta^3_{\mathbb{Z}/2}/\Theta^3_L\), The following corollaries are obtained. \textbf{Corollary E}. Let \(Y\) be an integral homology sphere obtained by \(1/n\) surgery on a knot \(K\) in \(S^3\) with \(n\) odd, and suppose that \(\mathrm{Arf}(K)=1\). Then \(Y\) has infinite order in \(\Theta^3_\mathbb{Z}\) if \(K\) is either of figure-eight knot, or a quasialternating knot with nonzero signature \(\sigma(K)\). \textbf{Corollary F}. Any knot \(K\) whose double branched cover satisfies the conditions of Theorem D generates an infinite syclic subgroup in \(C_s/C_{\mathrm{thin}}\). Here \(C_s\) is the smooth knot concordance group and \(C_{\mathrm{thin}}\) is its subgroup generated by the Khovanov-homology thin knots. Theorems C and D are proved in \S5. Before proving Theorem A, a brief review of the definitions and properties of \(\lambda_{\mathrm{SW}}(X)\) and monopole Floer homology (which is necessary for the definition of \(h(X)\)) is given in \S2. To extend the construction of the correction term in [\textit{T. Mrowka} et al., loc. cit.] to a rational homology \(S^1\times S^3\) manifold \(X\), the authors need to consider manifolds with ``long neck'' \(X_R\), where \(R\) is the ``length'' of the neck, with a metric of the form \(g=dt^2+h\) in the product region \([-\epsilon,\epsilon]\times Y\), that provides that the associated Dirac operator is invertible. The existence of such a metric is proved in \S10. Before actually proving the existence, assuming the existence of such a metric, the authors construct the correction term for a rational homology \(S^1\times S^3\) manifold \(X\) in \S6. The proof of Theorem A is reduced to the following Theorems 3.2 and 3.3: \textbf{Theorem 3.2}. Assume \(X_R\) has a metric satisfying the assumption above, then for all sufficiently large \(R\), the correction term \(w(X_R, g_R)\) is well defined and we have the equality \[ w(X_R, g_R)=n(Y,h,\mathfrak{s}). \] Here \(n(Y,h,\mathfrak{s})=\operatorname{ind}D^+(Z_\infty)+\frac{1}{2}\operatorname{sign}Z\), where \(Z\) is a compact spin manifold with spin boundary, \(D^+(Z_\infty)\) is the spin Dirac operator on the manifold \(Z_\infty\) with cylindrical end obtained setting \(X=S^1\times Y\). \textbf{Theorem 3.3}. Under the same assumptions as in Theorem 3.2, for all sufficiently large \(R\) and sufficiently small perturbations \(\hat{\mathfrak{p}}_R\) defined in \S8.2, which make the moduli space \(\mathcal{M}(X_R,g_R,\hat{\mathfrak{p}}_R)\) regular, we have the equality \[ \sharp\mathcal{M}(X_R,g_R,\hat{\mathfrak{p}}_R)=-\operatorname{Lef}(W_\ast:C^0\to C^0). \] Theorem 3.2 is proved in \S6 (Theorem 6.1). In \S11, the last section, an alternative proof via periodic \(\eta\)-invariants is given (Theorem 11.1, cf. [\textit{T. Mrowka} et al., Compos. Math. 152, No. 2, 399--444 (2016; Zbl 1348.58012)]). \S7 is devoted to estimating the first eigenvalue of \(D^-(X_R)D^+(X_R)\) as \(R\to \infty\). It is used to obtain a compactness theorem (Theorem 8.2) in \S8. In \S9, the authors prove a gluing theorem (Theorem 9.1). This asserts the existence a homomorphism \(\rho:\mathcal{M}(X_R)\to \bigcup_{[\alpha]\in\mathfrak{C}^\ast}\mathcal{M}(W_\infty,[\alpha])\), \(\mathfrak{C}^\ast\) is the set of critical points of the Chern-Simons-Dirac functional (cf. \S2.2). Theorem 3.3 is proved from the compactness theorem and the gluing theorem in \S9.4 (cf. [Zbl 1238.57028]).